How predictable is local erosion rate in eroding landscapes?

The current suite of numerical landscape models suggest that, under steady forcing, erosional landscapes evolve to a static steady state in which erosion everywhere balances uplift. Among other things, this implies that the only limitation on our ability to predict the future configuration of a landscape is imperfect knowledge of initial conditions and stochastic forcing events (e.g. storms, earthquakes). These are formidable obstacles to prediction, but they are apparently not the only ones. We have constructed a physical model of a drainage basin which erodes through several units of relief. We conducted several constantly forced runs at various base level fall and rainfall rates. The landscapes develop 3^rd to 5^th order stream networks, and erode by surface runoff, hillslope failures, and upstream migrating knickpoints. Within the constraints of an overall balance between uplift and erosion, interactions between streams and hillslopes result in spatially and temporally variable erosion rates. These results suggest that eroding drainage basins at steady forcing are intrinsically dynamic structures. Current numerical models do not exhibit the same level of erosional variability at steady forcing, suggesting that some feedback mechanisms may be missing from model formulations. The presence of inherent dynamism in eroding landscapes could seriously complicate predictions of local erosion rate, even if an average balance between uplift and erosion rate has been attained for a given drainage basin.